Coddy CLI

General-purpose AI agent in one Go binary: console, ACP editors, web UI, Telegram, a cron scheduler and MCP support

199 ★ on the repository
Last commit Nov 5, 2026
MIT

coddy-project/coddy-agentAuthor:coddy-project

Install

Works in: Coddy
Services: Pachca

Manual install

curl -fsSL https://coddy.dev/install.sh | bash

For Linux and macOS. The script puts coddy on PATH and creates ~/.coddy/config.yaml. On Windows the README gives irm https://coddy.dev/install.ps1 | iex.

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

The agent runs shell commands and edits files, and in agent mode it has the full toolset
The Telegram and Pachca gateways and the web UI expose control of the agent from outside, with rights set by access settings
The Coddy for Browser extension gives the agent browser tabs with your sessions, with approval requested per site

This is third-party code. Review the repository files before installing.

What it does

Coddy builds into one static Go binary and exposes the same agent loop through several entry points. The console TUI starts with coddy, the web UI and an OpenAI-compatible API come from coddy serve, and Zed or VS Code connect over the Agent Client Protocol with coddy acp. It also has a Telegram gateway, a Pachca messenger gateway, a cron scheduler with jobs written in Markdown and a remote mode for driving another node. All entry points share the sessions under ~/.coddy, so a chat started in one place continues in the browser. The agent has three modes (agent, plan, ask), rules from .coddy/rules, .claude/rules and .cursor/rules, SKILL.md skills, subagents, hooks in Claude Code's format and MCP servers over stdio, HTTP and SSE. Six skills ship inside the binary, including /crossreview for code review through external CLI agents. The Coddy for Browser extension hands the agent tabs of Chrome, Yandex Browser or Firefox after you approve each site.

Who it is for. People who want one self-hosted agent on a laptop, a server or in a container, reachable from the terminal, editor, browser and Telegram at once.

Good fit when

The agent should run on a server and answer in a Telegram bot while you continue the same conversation in a console on a laptop
You want your own OpenAI-compatible endpoint on top of any local model through Ollama or llama.cpp
Zed or VS Code should get an agent over ACP with your rules, skills and modes
Daily jobs such as a repository digest run on a schedule from a Markdown file
A fleet of agent containers should be observed from one place through the swarm relay and Prometheus metrics

Not a fit when

You only need an editor assistant without a server, messengers or a scheduler: a narrower CLI is enough for that
Your organization forbids piping curl into bash: the README offers such an installer, but deb and rpm packages and a Docker image exist
A Claude subscription is your main sign-in: the docs describe an API key for Anthropic, and account sign-in is documented for other providers

Example request

Read the README and project layout, draft a refactoring plan for the payment module and save it to docs/plan.md without changing any code

Limitations

A model is required: a key for any OpenAI-compatible provider, an Anthropic key, or a local Ollama or llama.cpp server.
The /login command signs in through ChatGPT via the Codex backend, a Devin account and OpenCode: check those services' terms yourself.
Browser features work only through the Coddy for Browser extension and after you approve each site.
Some features are compiled in by build tags (http, ui, scheduler, gateway, swarm, browser); the prebuilt releases ship the default set.
The project changes fast, with commits almost daily.

How to disable. Stop the service with coddy serve uninstall (Linux) or coddy serve stop, then delete the binary and the ~/.coddy directory with sessions and config. A Homebrew install is removed with brew uninstall coddy.

FAQ

How do I install Coddy on macOS?

With brew install coddy-project/coddy-agent/coddy from the project's own tap. Updates come through brew upgrade or coddy update -y. Linux has a script plus deb and rpm packages, Windows has a PowerShell installer.

Which models work with Coddy?

Any OpenAI-compatible provider, Anthropic, Ollama and llama.cpp. Providers and models go in ~/.coddy/config.yaml, and coddy -t validates the file.

How do I connect Coddy to Zed or VS Code?

The editor launches coddy acp as an ACP agent. Coddy's modes, models, permissions and skills then appear in its composer. Steps for each editor are in the Editors section of the docs.

Can I reuse Claude Code skills?

Yes, SKILL.md skills become slash commands. Rules from .claude/rules, hooks in Claude Code's format and MCP servers from mcp.json are read too.

Where does Coddy keep sessions and settings?

In the ~/.coddy directory: it holds config.yaml, shared rules and the sessions of every entry point. coddy -t checks the config and coddy docs opens the built-in documentation offline.

Discussion

No questions or reviews yet.

All Pachca tools

Pachca API skills and CLI

Пачка API

Official

Official agent skills and CLI for the Pachca corporate messenger: messages, chats, users, bots, tasks and search

SkillBroad access11

Pachca MCP

mcp-pachca

Russian stack

MCP server for the Pachca corporate messenger: chats, message search, users and sending messages

MCP serverBroad access4

sudocode

A memory and coordination layer for coding agents: specs and issues live in git while agents work in parallel

CLIBroad access296

Gentle-AI

A Go CLI that configures the AI coding agents you already use: it gives them shared memory, a workflow and deterministic steps without replacing the agents them

CLIExtended access8.5K

CoreCoder

A teaching terminal coding agent in Python, built as a "nanoGPT for agents": read it, take it apart and fork it

CLIBroad access1.9K

Codync

Codync turns Claude Code, Codex, Cursor and other agents on your computer into bots you message from an iPhone or a terminal